AEROSPACE EXPERT SYSTEMS 303

AEROSPACE EXPERT SYSTEMS

Since the Wright brothers first took flight, the progress of
mankind in the domain of flight has been nothing short of
spectacular. The progress of technology, faster aircraft, in-
strument flight, and increased air traffic resulted in establish-
ing a governmental infrastructure to control air traffic. Dur-
ing the late 1920s air traffic control was established as a
profession. Rules, procedures, standards, and facilities were
designed to provide for safe and orderly traffic patterns and
required separation between the aircraft (1). Stakeholders of
the airspace system include airline operators, airport authori-
ties, weather specialists, and air traffic controllers (ground,
tower, en-route, flow), as well as the passengers. In addition
to civilian users, including the airlines and general aviation,
the system must accommodate military and space flight activ-
ities.
The demand for air transport has steadily increased since
introduction of jet aircraft during the late 1950s. The safety
and effectiveness of the national airspace system (NAS) de-
pends on the performance of the air traffic management
(ATM) personnel—the employees of the Federal Aviation Ad-
ministration (FAA). The increasing complexity of the system
and proliferation of computing equipment have generated an
urgent need to explore the possibility of supporting the hu-
man component of the system with tools and techniques
based on the concepts and methods of artificial intelligence
(AI) (2).
Intelligence is defined as ‘‘the ability to learn or under-
stand from experience, ability to acquire and retain knowl-
edge’’ (3). Applied AI is about programming computers to per-
form tasks previously assumed to require human intelligence.
The usefulness of AI is the measure of its success. The key
issue is realization that knowledge must be represented ex-
plicitly in terms of its nonalgorithmic contents. The computer
program acts upon it by deduction and reasoning applying
various search algorithms. There is a need to create software
products representing an artificial expertise—a container for
limited-domain knowledge. This development is particularly
important in the case when the pool of available experts is
limited (or about to be limited in the future).
As has been noted in other works, ‘‘an AI system must be
capable of doing three things: (a) store knowledge; (b) apply
the knowledge stored to solve problems; and (c) acquire new
knowledge through experience. An AI system has three key
components: representation, reasoning, and learning’’ (3a).

ARTIFICIAL INTELLIGENCE—CONCEPTS AND APPROACHES

The difficulties associated with learning have led people to

use other methods for augmenting knowledge bases. Expert

J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
304 AEROSPACE EXPERT SYSTEMS
systems (ES) are such an example, where the compiled knowl-
edge and experience of a human expert are used in lieu of
having the system develop its own experience, duplicating
that of the expert. This is not to say that learning cannot be
part of an ES.
The methods and techniques of AI are well suited for appli-
cations not amenable to standard, procedural, problem-solv-
ing techniques. Examples of such applications are where the
available information is uncertain, sometimes erroneous, and
often inconsistent. In such a case, using quantitative algorith-
mic calculations may not lead to the solution, whereas use of
plausible and logical reasoning may. The approach taken by
AI leads generally to a nonoptimal, but acceptable, solutions
reached by using rules of thumb and logical inferencing mech-
anisms.
For such an approach, the system is represented by a fac-
tual description in the form of chunks of meaningful data
(knowledge) related to the system state and by the relation-
ships among those data. An external, domain-independent in-
ferencing mechanism makes it possible to draw new conclu-
sions from existing knowledge resulting in changes and
updates of the knowledge base. The AI discipline concerned
with these issues is called knowledge representation. There
are various paradigms of how to represent human knowledge:
predicate calculus, production rules, frames and scripts, and
semantic networks. The selected representation scheme must
express all necessary information, support efficient execution
of the resulting computer code, and provide a natural scheme
for the user. AI is concerned with qualitative, rather than
quantitative, problem solving. Thus the selected knowledge
representation and the used tools must be able to (a) handle
qualitative knowledge, (b) allow new knowledge to be created
from a set of facts, (c) allow for representation applicable to
not only a specific situation but also to general principles, and
(d) capture complex semantic meaning and allow for meta-
level reasoning (reasoning about the knowledge itself, not just
the domain).
A distributed-intelligence system (DIS) is the concept of a
system operated by a machine and managed by a human. The
human operator is involved in planning, making decisions,
and performing high-level functions, whereas the machine
portion of the system executes most of the system’s regular
operational functions, collects and stores data, and handles
routine decision situations with a limited number of options
(4). Such an approach requires further research in the area of
man–machine interface and physiopsychological aspects re-
lated to the stress and anxiety factors.
Decision support systems (DSS) are computer systems that
advise human operators or are automated systems that make
decisions within a well-defined area. The systems are used
where similar decision processes are repeated, but where the
information to decide upon may differ. Some DSS are known
as expert systems (ES): they imitate human expert behavior.
Decision procedures of an expert are analyzed and trans-
formed into rules and subsequently implemented into the sys-
tem. The ES is a computer program providing solutions to
problems normally requiring a human expert with an appro-
priate domain knowledge and experience. The experts are em-
ployed to solve problems requiring planning or decision mak-
ing. They frequently use rules of thumb—heuristics based on
experience, analogies, and intuitive rationale to explain the
behavior associated with their area of expertise. Development
of ES requires identification of the thought process describing
how the human expert solves a specific problem. There are
three steps in this identification (1) knowledge representa-
tion, (2) knowledge acquisition, and (3) knowledge proces-
sing (5).
To imitate a human expert, a successful ES must reach the
solution based on the predefined body of knowledge, despite
incomplete or uncertain data; explain and justify how the so-
lution was reached; and communicate with the user and/or
environment, thus acquiring new expertise. Some other ad-
vanced properties include the ability of natural language com-
munication to support an effective user interface and easy
knowledge acquisition. The ability to determine relevance, by
either giving a referral or degrading gracefully at the exper-
tise boundary, is another important human characteristic, as
are common sense reasoning and breaking rules. The issue of
effective acquisition of new knowledge and reorganization of
expert knowledge base is closely related to machine learn-
ing—another important AI discipline closely associated with
cognitive science.
Besides the aerospace industry, other successful imple-
mentations of ES in such areas as medical diagnosis, manu-
facturing, and military command and control have shown po-
tential for further expansion. However, the analysis of such
systems is, in most cases, time-neutral. That is, most expert
systems are not able to look into the future and predict future
system behavior. Reasoning on a given problem is applied in
the current time instant. Only an integrated DSS, combining
the potential of simulation and the reasoning ability of ES,
can be used in time-dependent applications (6).
Various knowledge acquisition schemes have been devel-
oped to extract expert knowledge. Interview techniques and
observation of humans performing expert tasks have been
used to elicit knowledge. If the problem domain is well de-
fined, the approach leads to a rule-based system, where the
expertise is represented as a set of if–then rules. An inferenc-
ing mechanism searches the rule base, drawing inferences
and asserting/retracting the facts about the problem at hand.
For the expert operating a complex system, it is easier to ex-
plain general experiences rather than identify specific rules.
This observation led to emergence of case-based reasoning,
where various operational cases are acquired and modeled
into the system knowledge base. Using the process of induc-
tion the ES infers new rules based on the current facts and
the experiences from previous similar cases. The system then
produces advice and modifies its knowledge base (7).
Uncertainty is an inseparable part of any decision process.
It may be caused by insufficient understanding of the problem
area, by missing or uncertain measurements and observa-
tions, or by nondeterministic causal relations. The emergence
of Bayesian networks allows researchers to model the prob-
lem area with its built-in uncertainties. This approach can
result in a directed graph, where the vertices represent the
variables of the problem area, and a directed edge between
two vertices means the state of one variable influences the
state of the other. The magnitude of this influence is repre-
sented by a conditional probability (8). Another approach to
uncertainty is the application of fuzzy set theory, which
allows researchers to assign a range of quantitative variables
to create qualitative entities subsequently handled by the
ES (9).

NATIONAL AIRSPACE SYSTEM
The FAA has been modernizing the NAS in a massive project
initiated in the early 1980s. Several papers (see reading list)
related to the application of AI methods and techniques to air
traffic control (ATC) in the NAS of the twenty-first century
describe attempts to formalize the ATC knowledge base and
build systems assisting controller operations and training.
Most of the systems use a simulation component designed to
predict the future state of the system.
The FAA provides ATC and the traffic flow management
(TFM) functions, collectively referred to as ATM, designed to
maximize air traffic throughput and minimize delays while
maintaining a safe operating environment. The overall ATM
system is designed to give equitable access for all NAS users,
while ensuring safe separation of aircraft from each other and
from terrain/physical obstacles and restricted airspace. Physi-
cal and operational constraints of the NAS include availabil-
ity of airport runways, severe weather conditions, equipment
outage, and heavy traffic demands. These constraints limit
the options to accommodate user preferences for flight times,
routes, and altitudes.
TFM is the ATM function that focuses on managing NAS
resources, mitigating contention for scarce capacity, and dis-
seminating information about the anticipated restrictions and
demand predictions. TFM is intended to allocate capacity to
NAS users in an equitable fashion and allow them to main-
tain operational control.
The ATC functions of the ATM include real-time separa-
tion and sequencing of aircraft en-route, during arrival and
departure at the terminal, and ground control on the airport
surface. En-route controllers in 20 centers of the continental
United States provide services for aircraft between the depar-
ture and arrival phases of flight at air route traffic control
centers. Services include separating aircraft, monitoring traf-
fic flow, implementing traffic management initiatives pro-
vided by TFM, issuing traffic and weather advisories, coordi-
nating special use airspace, and providing emergency
assistance. The terminal controllers in more than 50 larger
metropolitan airport areas provide ATC services to aircraft
traveling between an airport ground control and the en-route
environment. The terminal area includes one or more airports
controlled out of a terminal radar approach-control facility.
The services include aircraft separation, sequencing, traffic
advisories, and alerts, significant weather advisories, and ra-
dar vectoring for arriving, departing, and through traffic. Air-
port ground control is responsible for all vehicles operating on
taxiways, aprons, and gate areas.
Advanced ATM functionality is being developed to support
real-time information exchange between ATC controllers and
airlines, facilitating collaborative decision making among air-
space users. Advanced ATC decision support tools are also
being developed to increase safety, efficiency, and flexibility
as the NAS system evolves toward free flight. These advanced
functions must be integrated into the flow, en-route, terminal,
and tower/ground facilities as part of a coherent national sys-
tem in order to deliver their full benefits (see Web resources).
AIR TRAFFIC FLOW MANAGEMENT OPERATIONS
The Air Traffic Control System Command Center (ATCSCC)
is located in Herndon, Virginia. The current role of the center
AEROSPACE EXPERT SYSTEMS 305
is to manage the allocation of NAS resources and limit air-
borne delays. These objectives are accomplished by imple-
menting TFM initiatives: ground delay program (GDP),
ground stop program (GSP), miles/minutes-in-trail restriction
(MITR), and severe weather avoidance program (SWAP). The
center is staffed by experienced specialists with an extensive
knowledge of ATC procedures and familiar with the impact
of weather conditions and possible TFM initiatives on NAS
performance (10).
There are a wide variety of computer tools based on avail-
able aircraft data to support the specialist operations. Data
can be displayed in both character and graphic formats, show-
ing, for instance, all aircraft scheduled in a specific sector
within a specific timeframe, or all aircraft scheduled to arrive
at a specific airport. The hourly arrival demand for an indi-
vidual airport may be displayed and printed. Weather condi-
tions are displayed graphically, including areas of limited
ceiling and visibility, precipitation, expected thunderstorms,
and jet streams. There is easy access to alphanumeric local
ground weather reports, altitude profiles, and briefings from
radar observation.
The traffic management specialist responds to the weather
situation and to requests from the major airports in the clus-
ter. In cases when an airport acceptance rate is anticipated
to decline (deteriorating weather conditions, airport configu-
ration change) the flow controller may consider implementa-
tion of the GDP for that airport. The program can be imple-
mented for any combination of the en-route centers, from
adjacent to the entire system. The scope of the program, in
terms of duration and affected areas, is based on the current
situation and determined as the result of the controller’s
knowledge and experience. The GPD software recomputes de-
parture times, and estimates predicted delays. When the com-
putation predicts acceptable delays in the system, the special-
ist sends the new schedule to the centers and the airlines for
implementation. In the case when an airport is unable to op-
erate or experiences severely reduced capacity with already
long delays and surplus traffic, the specialist may order the
GSP for the flights destined to the affected airport. Both GDP
and GSP affect only the aircraft scheduled for later departure.
Any action is coordinated with all interested parties before
implementation. The shift supervisor has the final authority
on whether or not the proposed plans are implemented.
A regional traffic management unit may request MITR in
cases of reduced acceptance rate of the arrival sector caused
by weather, traffic volume, or staffing problems. The situation
is analyzed and coordinated by the area cluster specialist, and
the outcome is conveyed to the affected centers. The role of
flow control is limited to a mediation between two adjacent
centers.
Severe weather conditions en-route may force a center to
request more forceful measures as a significant rerouting of
traffic. A separate cluster of flow personnel manages the im-
plementation of SWAP rerouting. The position is equipped
with an additional workstation with a database of airport con-
figurations under different weather conditions and the prefer-
ential routes among all major airports. The main role is to
provide coordination for the new routing.
There is a significant amount of domain knowledge in-
volved in TFM activities. For example, some of the airports
require implementation of a GDP/GSP for the entire system,
whereas others may restrict the program to the adjacent cen-
306 AEROSPACE EXPERT SYSTEMS
ters only, based on incoming traffic patterns and past system
responses. Rerouting must consider the capacity of the af-
fected sector to handle the additional traffic. Decisions result
from the anticipated situation and knowledge of the system
operation and its behavior. Therefore, most decisions are
based on the flow specialist’s personal experience.
Expert Systems for Air Traffic Flow Management
The FAA Operations Research Service developed SMARTFLO—
a knowledge based system. This prototype computer program
provides the TFM specialists with a user-friendly support tool
capable of suggesting a set of required traffic management
initiatives for a specified time horizon (11). The system rea-
soning capability was based on the available information (in-
quiring for more information if necessary), and the base for
decision and the line of reasoning and was presented to the
user. The system includes encyclopedic knowledge about
NAS operations.
The basic activities in ES development are knowledge rep-
resentation, knowledge acquisition, and feasibility prototype.
The TFM system is described in terms of its concepts and
activities. Among the concepts defined are pacing airport, af-
fected area, weather, volume/demand, TFM initiatives, pre-
ferred route, rerouting, and arrival distribution to the airport.
TFM activities include acquiring information about the sys-
tem and weather and subsequently processing the informa-
tion generating the required TFM initiatives. A GDP initia-
tive is proposed, using heuristics based on the flow
controller’s knowledge. The basis is the same: The action is
considered when anticipated demand exceeds the expected ca-
pacity. However, the numerical values used (e.g., rate, time
interval), the area involved, or freedom to use other options
(GSP, MITR) are left to the specialist.
The initial knowledge-acquisition phase included observa-
tion of daily operations and interviews with ATCSCC person-
nel. The analysis of daily reports including specialist logs,
weather reports, listing of traffic management initiatives for
the day, and estimates of system performance (delays) pro-
vided an additional source of knowledge. As the result of
these activities, a knowledge representation scheme (facts
and rules) was selected and a prototype was implemented.
CLIPS, an ES tool developed by NASA, was used as the infer-
encing mechanism. Subsequently, the Sun workstation-based
graphic interface was added to complete the initial implemen-
tation. The advanced prototype of the system, implemented
on a Hewlett Packard HP9000 workstation, used TAE+
graphic interface tool with active agents and the object-ori-
ented paradigm. The objects contain information about their
relationships with other objects, identification, source infor-
mation, definition and description, lists of relations, rules,
and constraints defining their behavior. The classes of object
in SMARTFLO are jobs, agents, resources, operations, models,
events, systems, devices, sites, and environment.
The SMARTFLO advanced prototype graphic windows allow
the user to select the airport/center, enter data, query for sys-
tem information, and watch suggested TFM initiatives being
generated. A user-friendly consultation allows the user to ask
‘‘What should we do?’’, and ‘‘Why?’’. In addition to generating
the suggested TFM initiatives and playing what-if scenarios,
the system includes an electronic encyclopedia with extensive
information about ATCSCC operations.
INTELLIGENT TRAINING
Computer-aided instruction (CAI) is a common use of comput-
ers in education and training. CAI tools incorporate well-pre-
pared course materials and lessons plans into routines opti-
mized for each student. However, conventional CAI tools are
limited to either electronic page-turners or drill-and-practice
monitors, severely limiting the overall effectiveness of the
system in a situation where declarative knowledge is sought.
The incorporation of AI techniques into CAI spawned the
creation of intelligent tutoring systems (ITS) capable of mod-
eling the student learning process, drawing conclusions from
student problem solving behavior, and modifying the se-
quence in which material is presented to the student (12). ITS
is intended to help individual students identify their specific
weaknesses and rectify them effectively and to be sensitive to
the student’s preferred style of learning. The objective of some
researchers is to produce entirely autonomous ITS based on
pedagogical expertise and the principles in the domain
knowledge.
The major blocks of a modern simulation-based ITS are (a)
simulator, (b) domain expert, (c) student model, (d) evaluator,
(e) scenario generator, (f) training manager, and (g) user in-
terface. The simulator represents the real-life system for
which the student is being trained. The domain expert con-
tains the body of knowledge that should be presented and
taught to the student. It is also used for evaluation of student
performance and the overall learning progress. To achieve
these objectives, most systems generate and store all feasible
solutions to the problems in the same context as the student,
so that their respective answers can be compared. The stu-
dent model contains knowledge about the student’s under-
standing of the material. This knowledge is extremely impor-
tant in the decision making process affecting the choice of
subsequent tutoring strategies. The evaluation module is
used to evaluate the student performance based on the situa-
tion assessment derived from the simulation status. The sce-
nario generator is used to generate realistic training scenar-
ios appropriate for the student. The core of the system is the
training manager, containing the knowledge about teaching
methods. The training manager, based on the current evalua-
tion, selects the next scenario component from the scenario
generator monitoring the student’s performance. The module
uses a decision making process based on teacher experience
and assessment of past training sessions. Finally, the success
of an ITS depends significantly on user-interface quality. All
of the components are coordinated by a communication entity
referred to as a blackboard. Any ITS module can place infor-
mation on the blackboard making it available to all other
modules.
An ITS is often in the form of a computer-based problem
solving tutor, a coach, a laboratory instructor, or a consultant.
For the development of an ITS for aerospace (ATC specialists,
pilots, astronauts, and airline dispatchers), the most suitable
tutoring strategy seems to be a combination of coaching and
guided-discovery learning. The student is in full control of the
activity for which the tutoring is provided—a simulated ver-
sion of a real system. Simulation is used because it provides
an attractive motivational context for discovery learning. The
coaching task of the ITS is to foster the learning inherent in
the activity itself by emphasizing existing learning opportuni-
ties and by transforming failures into learning experiences.

The potential of coaching is evident when ITS is embedded in
an interactive, real-time simulation. It must be noted that
many simulation-based ITS are strong in their simulation
capabilities, but rather weak in delivering an effective tu-
toring strategy.
Air Traffic Control Intelligent Training System
The ATC-ITS prototype system (13) was designed to support
ATC training by providing the student with an interactive,
simulated ATC domain by creating effective training scenar-
ios with exercises matching the student performance level, by
coaching the student through the scenarios, and by evaluat-
ing student performance.
The first major task is to build a knowledge base repre-
senting the domain expertise of an en-route air traffic control-
ler. The principal methods used are observations, informal
and formal interviews, operational task analysis, and simula-
tions. A key element is the advice provided by air traffic con-
trollers with both operational experience and en-route train-
ing expertise (14). One aspect of ITS development focuses on
the expert system and its interface with a realistic simula-
tion. The other aspect deals with the training methods, con-
troller characteristics, student capabilities and skills, devel-
opment of training strategies, and assessment techniques for
student performance evaluation.
The ATC-ITS provides training for en-route ATC students.
It is assumed that the trainee has control over aircraft within
a specified training sector. The trainee is responsible for
hands-off procedures including receiving incoming and trans-
ferring outgoing aircraft. The transfers and clearances must
conform to the sector standard operating procedures. The
route through the sector must closely follow the flight plan.
The students must conform to ATC communication standards
and phraseology, secure an efficient and orderly movement of
aircraft through the sector, and provide appropriate re-
sponses to pilot requests. The trainee is primarily respon-
sible for ensuring proper aircraft separation within the con-
trolled sector. All the activities are monitored and evaluated
by the ITS.
The simulation, implemented in object-oriented MODSIM
language, provides a realistic, real-time, en-route air traffic
sector simulation displaying aircraft targets with data blocks,
flight paths, and flight strips, as well as controller-to-pilot and
controller-to-controller communications. ATC-ITS graphics
represents an en-route radar scope. It displays sector bound-
aries, airways, fixes, restricted areas, targets, data blocks in
different forms (limited, full, separation violation, handoff ini-
tiated, handoff accepted, radio failure, emergency, etc.). The
graphics display responds to mouse operations such as hand-
off initiation, handoff acceptance, data block move, and alti-
tude or route change for selected aircraft. The communication
window simulates the radio communication between the con-
troller and simulated pilots.
The air traffic expert controller is an implementation of a
rule-based ATC expert model. It can provide expert knowl-
edge relative to ATC decision making. The knowledge base
consists of facts and rules. The situation facts are assessed
from the simulation module and the user input. The facts rep-
resent aircraft status, sector situation, controller workload,
pilot and adjacent controller requests, emergency events, and
so on. The action facts reflect the required action to be imple-
AEROSPACE EXPERT SYSTEMS 307
mented by the controller as a direct response to the given
situation. The rules reflect the controller experience in tan-
dem with ATC operational procedures and regulations. The
system uses a forward-chaining mechanism with its produc-
tion rules where the situation facts are the rule premises and
the action facts are the rule conclusion. The CLIPS inferenc-
ing mechanism uses the Rete algorithm (15) to search the rule
base. Assertion or retraction of the facts to or from the knowl-
edge base constitutes the principal knowledge control han-
dling mechanism (16). The ATC rules represented in the
knowledge base are used by CLIPS inferencing to arrive at
the ‘‘best’’ decision. Firing the rules leaves a trace, which pro-
vides the user an explanation about the line of reasoning
leading to the specific recommended action. The prototype
system was developed on a Sun workstation with a Sunview
windowing system.
Expert Systems for Pilots
ES in aerospace are often designed to aid the operator of the
system, including piloting vehicles. Typically the use of an ES
in an operational environment such as piloting can be ori-
ented to supporting the pilot or assessing the pilot. Often, ES
that are capable of assessment incorporate some form of feed-
back or critique to the pilot, further enhancing their use-
fulness.
Pilots perform three fundamental activities while flying.
The first, and primary task, is control. This relates directly to
the pilot’s ability to manage the attitude of the aircraft about
the three axes of the aircraft. ES that support control include
stability and inertial reference systems. The second activity
the pilot performs is navigation. This is control of the move-
ment of the aircraft from point A to point B. Pilots are sup-
ported by ES such as autopilots and navigation computers.
The third activity of the pilot is the need to communicate in-
formation relating to the aircraft. ES can provide data link
and information fusion based on the pilot’s current and future
activities. Knowledge acquisition for these types of systems
can be very difficult, costly, and time consuming. Techniques
to improve the process include multiple model integration, in-
dexing, and multiple input summarization. Examples of sys-
tems supporting the pilot are Pilot Associate (17) and Hazard
Monitor (18). Both systems are based on real-time data acqui-
sition from the aircraft avionics. The system processes the
data to create knowledge chunks, used as knowledge base
by ES.
Control
Pilots generally understand that aircraft are dynamically un-
stable. This is true, even when tendencies toward instability
are rendered invisible by avionics designed to override those
innate problems. As a result, the human ability to control air-
craft attitude is often suspected in control-related incidents.
Computing systems supported by expert knowledge about the
engineering and dynamics of the vehicle are often employed
in these situations. While these may not be purely and exclu-
sively expert systems, they provide an excellent, if somewhat
oversimplified, example. Flight stability ES use a theoretical
approach, coupled with a dynamic input, to create a response
algorithm. The use of an optimal control model provides the
system with a baseline of engineered performance. The ES
308 AEROSPACE EXPERT SYSTEMS
designer would typically use a rule-based approach to imple-
ment this model. Dynamic input is provided by both the pilot
and the flight computers and by inertial reference computers
(if available), allowing a control model to be exercised. Often,
a sophisticated formula or Bayesian loop is used to control
the limits of the autopilot system. Engineers are concerned
with the pilot’s induction of out-of-phase or undampened in-
put to the system, and thus, in some aircraft, such as those
produced by Airbus Industrie, the autopilot system will actu-
ally retain and enforce control input limits made by the pilot.
Navigation
Navigating an aircraft is sometimes simple and sometimes
complex. As with control, a number of ES-type technologies
exist to ensure that the task of navigating the aircraft is safe
and effective. When coupled with the control computers, the
autopilot can take on the task of determining position and
guiding the aircraft to its intended destination by returning
commands to the autopilot system. This would imply that air-
craft are comprised of complex ES networks, which is the case
in many advanced turbojet and most military aircraft. Some
corporate and general aviation aircraft have similar although
less expert systems.
When a pilot of any aircraft is operating in conditions that
are favorable for seeing and navigating the aircraft by looking
out the window, the use of such systems is less critical. In
these circumstances, ES are often employed to monitor and
assess pilot navigation performance rather than actively navi-
gate the aircraft. In conditions where the pilot relies heavily
on ES to navigate the aircraft, such systems are designed to
mimic the decision process and control input that would be
provided by the pilot. It is common to have an ES that uses
both inductive and deductive structures to provide the input
to the control algorithms. Most often the system design is re-
duced to having a known position and a set of control rules
derived for the primary task of controlling (discussed pre-
viously) which combine to create a desired effect—the future
location. This process is implemented in an ES using deduc-
tive reasoning.
Communication
When exploring pilot communication activities, a number of
different communications take place where ES are employed.
Pilots receive information from the aircraft systems in the
form of displays, and send information to each other and to
others on the ground. A remarkably clear pattern-of-informa-
tion needs exists during a large percentage of the time pilots
are flying. Using this pattern, ES designers have imple-
mented systems that anticipate and provide the information
needed when it is needed. Typical systems include the auto-
mated information and crew alerting systems used to monitor
aircraft systems, detect trends and anomalies in the system,
and alert the crew to the problem. These are truly ES, in that
they gather data and, rather than merely responding to it,
they analyze it, consider alternative responses, and then initi-
ate action.
These ES are found on most transport and military aircraft
and are developed using engineering data to derive functional
limits, which in turn support both rule-based and input-
driven algorithms. Other forms of ES, which support pilots by
managing information, are used to communicate data from
the ground to the aircraft. Based on a set of rules developed
by industry, a priority is assigned and managed to ensure the
timely, accurate, and intelligible delivery of data between the
ground station and the aircraft. These ES function on the
premise that navigation and environmental information, such
as weather reports, may or may not be relevant to the partic-
ular aircraft receiving the broadcast. Using rules and real-
time information residing in the aircraft itself, the ES moni-
tors, acquires, and posts information for the pilot based on
the criticality of the information already in the queue.
Pilot Knowledge Acquisition
Methods of capturing knowledge of both control and naviga-
tion activities are varied. Certain things are known about the
aircraft that are derived from the engineering process. Other
systemic effects, such as the role of the pilot, are less certain,
and there is a need to capture expert knowledge. One such
method is the use of models that present reasonable represen-
tations of the expert.
Multiple model integration is used to reduce the need to
explicitly define the knowledge for all cases and creates spe-
cific rules that fire in general conditions (environment is de-
fined heuristically). This method employs both detailed and
general knowledge acquisition and modeling, while yielding
high confidence in the rules that fire. Piloting is well suited
for such implementations, because the use of procedural
knowledge to induce rules can be used to meet the need for
specificity, whereas the general environmental conditions
may be described using generalizations. The use of concept
mapping is another method of reducing knowledge acquisition
problems in complex situations. Concept mapping allows spe-
cialized knowledge in the form of heuristics and problem-solv-
ing methods to be explicitly associated by the knowledge us-
ers with static facts and general knowledge (19). Specific to
concept mapping is the use of two techniques: first, the use of
combing multiple input whereby the experts have collectively
generated a summary map of the knowledge required for the
particular domain; the second technique is that of indexing,
which results in the development of themes and key concepts
that emerge from the relationships generated in the summary
mapping process.
Evaluation of Pilot Performance
ESs most often emerge as the result of capturing knowledge
of one or more experts in a format that facilitates computa-
tional analysis. Typically the analysis is focused on directing
the system to carry out certain rules and implement routines
that have been programmed into the system. Often the goal
of the ES designer is to achieve a human-like behavior from
the system. One of the more difficult tasks of the ES designer
is to implement a system that, in real-time or immediately
following use, can assess the performance of the human com-
ponent of the system. This is different from simply developing
an ES that can influence the operator during its use. The ES
needed to critique human behavior is required to have a set
of analysis capabilities that not only relate to the rules and
networks used in the system, but a set of principles that have
their roots in the psychology of human actions.
A number of systems exist that provide such an ability to
assess pilot performance. These systems are developed to aid
in predicting human performance, critique performance, and

quantify performance for use in a variety of future activities
such as curriculum design, monitoring and controlling qual-
ity, or systems engineering. Conducting knowledge acquisi-
tion for assessment requires the classification of the ‘‘expert’’
relative to the system user.
Pilots operate in a dynamic environment and the variables
that comprise the reasoning and observable behavior of pilots
is often very abstract or difficult to define. One of the best
examples of an ES used in assessing pilot performance is the
U.S. Navy effort to design and implement a model of the pilot
in flight-test maneuvers. The pilot model is a combination of
skill-based models and rules-based models (multiple model in-
tegration) whereby the skill-based portion provides a feed-
back path and the rule-based portion provides the inference
and feed-forward path (20). This ES employs the use of the
Rasmussen’s commonly recognized model of human error
comprised of the hierarchical categories of knowledge-based,
rule-based, and skill-based behaviors thought to lead to hu-
man error. Using a quasi-linear model, the system is capable
of quantifying skill-based pilot behavior. To resolve discrete
decision tasks, the use of a fuzzy-logic scheme is employed,
which supports rule-based behavioral assessments and, when
combined with the skill-based models, results in inference
rules that can be used to derive large classes of pilot tasks.
Predicting Performance
In a complex system such as an aircraft, the need exists to
select the best possible match of pilot and vehicle. To do this,
ES are used in ground-based test environments to provide a
dynamic environment that adapts to the behavior of the pilot
and challenge the pilot based on the correctness of the prior
action. These systems utilize psychological principles of hu-
man attention and motor skill to create a multiprocessing re-
quirement in the human mind. ES capable of resolving these
challenging environments then operate in the background of
the test apparatus to provide a predictive model to the ex-
pected behavior of the pilot, who is often behind the ES in
resolving the problem. The predictive nature of these systems
is implemented using both rules-based and neural network
structures. The ES controls the timing and the difficulty of
the task based on a Bayesian process that involves the users
input and changing heuristic models that form the initial
foundation of the models. The ES also uses a series of time-
related rules that are derived as the system is exercised to
control the tempo of the primary task—flying the aircraft.
Secondary tasks are driven by the ES control algorithms and
the pilots input. Results are compared to the past historical
results of others who have been used to establish the baseline
for comparison.
Quantifying Performance
Quantifying performance can be aiding by using ES to stan-
dardize measurement and classification of pilot performance.
An ES used in quantification of pilots in training uses a rules-
based approach and an optimal model to infer a score on a
pilot’s ability to maneuver the aircraft, given a standard. This
standard is used to define the baseline (quantify the optimal
performance) and a simple network of all outcome collected in
the past performances is used to define the pilots expected
performance. Using a deductive process, the data collected in
the pilot’s maneuvering of the aircraft is then reduced to a set
AEROSPACE EXPERT SYSTEMS 309
of rules where the antecedent is compared with the optimal
performance antecedent and the analysis is performed based
on quantitative data. This greatly improves the ability of the
ES to acquire new data and weight the data according to the
best known performance of the pilot prior to the event under
study. The resulting analysis can drive a rules-based model
that will then narrow the set of variables, and identify the
rules that fired, leading to a refined set of maneuvers that
need attention in future training.
Another ES used to quantify performance is an induction
engine developed for studying accident data. It can be said
that most accidents involve pilot error. These accidents, when
reduced to data, which has been processed using a qualitative
approach, can lead to key rules derived through induction.
While this method does not result in immediate ES utility,
the rules are necessary for the construction of the network of
events that preceded the accident. In applying these rules to
the performance of the pilot in a simulated environment, the
ES is able to rank the pilot’s performance by quantifying the
rues which, in the past, had a high probability of firing if
certain pre-accident events were present.
Knowledge Acquisition Is Based on the Target User
Since the expertise of the user is a known factor in the suc-
cessful implementation of an ES, expertise is therefore rela-
tive to the design strategy. The modeling and design of the ES
must be in concert with the level of expertise to be assessed.
Building systems that can be flexible and adapt to such re-
quirements require the ES designer to apply elective automa-
tion design theories. Typically ES are limited to use by one of
three classes of users. The first, the novice, finds the most
system utility at a level where rote or procedural knowledge
is sufficient, engages the highest use of automation when
available, and seeks simplicity in the interface. Second, the
journeyman finds utility in the middle range of the system,
partially utilizing manual and automated functions, and will
tolerate most levels of complex interfaces with some training
and practice. Last, the expert is capable of self-instruction on
the system since they operate the system using principles and
experience-based knowledge. Experts will not tolerate limited
flexibility in the system’s functional implementation.
EXPERT SYSTEMS AND SIMULATION
IN OTHER AEROSPACE APPLICATIONS
The close relationship between aviation and simulation dates
from the very beginning of aviation. The widespread use of
digital computers and software-based simulations have
broadened the scope of aviation simulation. By analyzing
computer simulation in aviation and aerospace, we can iden-
tify the following main application areas: (a) flight dynamics
simulation, (b) flight simulators for training, (c) simulation of
air traffic, and (d) simulation of aerospace systems to support
control and decision making.
Computer simulation is a technique that provides for imi-
tation of real-life situations using computers. Models and sim-
ulation have been used to facilitate understanding of the real-
world phenomena. Computer simulation is the discipline of
designing and executing a model of an actual system with
the subsequent analysis of the simulation output (21). In the
complex world of aviation and aerospace, building the correct
310 AEROSPACE EXPERT SYSTEMS
model is essential for the simulation designer to have a good
grasp of the domain. Conversely, managers responsible for
aerospace projects requiring simulation should be aware of
the advantages and shortfalls of simulation techniques and
methods. A careful approach to the design, implementation,
verification, and validation of the model must be taken. The
simulation experiment must be well designed, and feedback
from the real-world system needs to be used as often as possi-
ble. Abundant research proposes using a knowledge-based ap-
proach to support the simulation.
There are many ways that the simulation and ES can be
combined: ES can be used as an intelligent front-end to simu-
lation. ES may invoke the simulation run to generate the re-
quired system data. Simulation may be used to update time-
related variables in the ES. Finally, simulation can be a tech-
nique for ES testing (22).
The AI techniques have enhanced the historically strong
relationship between aerospace training and simulation. Us-
ing AI may reduce the impact of subjective, impatient, over-
loaded, and error-prone human instructor links in the train-
ing process. Efforts are being made to design systems that
incorporate techniques of AI and use the full power of dy-
namic simulation, including animation and graphics. An ex-
tensive application of these techniques gives one possible an-
swer to the problem of effective training in situations
resembling real-world conditions.
Flight Dynamics Simulation
The main feature of the flight dynamics simulation is that the
aircraft model representing the handling characteristics of
the airframe, engines, and the control systems is encoded in
the computer. The flight dynamics simulation is based on a
rigorous quantitative mathematical model expressed in terms
of continuous differential equations. Research on interfacing
such quantitative simulation of the aircraft in flight with a
qualitative simulation, in an attempt to support decision
making, has been presented in (23). The system extracts
quantitative data from a mathematical model of aircraft flight
dynamics and uses fuzzy inductive reasoning on the qualita-
tive model to recognize the flight accidents.
Fuzzy Reasoning (or Fuzzy Logic) is based on the theory of
Fuzzy Sets pioneered by Zadeh (9). It extends the conven-
tional logic introducing the concept of partial truth—truth
values between ‘‘completely true’’ and ‘‘completely false.’’
Fuzzy Reasoning attempts to mirror the imprecision of the
real world by providing a model for human reasoning in
which even the truth is not an absolute but rather a matter
of degree. Fuzzy Logic has emerged as a key methodology in
the conception, design and deployment of intelligent systems.
Flight Simulators
The flight dynamics model and simulation is the first step for
creating a ground-based flight simulator, designed to repro-
duce the behavior of an aircraft in flight. The simple simula-
tors are static flight procedure trainers with the flight instru-
ments driven by the simulation engine. Medium-range
simulators use high-fidelity graphics representing a dynamic
out-of-the-window view with appropriate terrain database.
The high-end simulators use a six-dimensional mechanical
platform to imitate the motion cues of the flight. The trainee
operating the simulator controls experiences the responses of
the model, displayed on the instrument panel and graphic
screens similar to those in the real aircraft. Additional ele-
ments to be imitated are the communication system (radio),
environmental system (e.g., engine noises), weapon system (in
the case of military aircraft), etc. The multidisciplinary na-
ture of a flight simulator requires specialized design skills
and fundamental knowledge. The problems related to inte-
grating components representing aeronautical, mechanical,
electrical, and computer engineering are of primary signifi-
cance.
Simulation of Air Traffic
Discrete simulation of air traffic has been used extensively to
imitate flow of aircraft through the airspace. Applications
range from the movement of the aircraft on the ground to
the en-route and oceanic flow. The models describe aircraft
behavior in all phases of the flight. Using well-established
methods of discrete dynamic simulation, the air traffic models
can determine a range of system parameters such as time
delays, waiting queues, capacity, and occupancy. They may be
used to test different ATC procedures, airspace configuration,
airport development, and changes in traffic demand and pat-
terns. Combined with animated high-fidelity graphics, the
discrete simulation can be used also as a tool for training and
human-factor research. Such models are used as tools in air-
port planning and design in relation to both aircraft and pas-
senger flow.
Often considering individual aircraft is not practical in
aerospace strategic models geared toward TFM. The continu-
ous approach where the network of air routes is simulated
with flow rate in the links may be used. Techniques of simula-
tion and operation research are used to determine capacities
of airways, airports, location of hubs, airline schedules, etc.
They are used both for analysis and design as well as for the
actual TFM.
One of the comprehensive simulation systems is Total Air-
space Airport Modeler (TAAM), developed by The Preston
Group, combining real-time simulation and an interactive
graphic user interface with AI elements. The system supports
the construction of aircraft performance data, airport and air-
space elements, procedures and strategies, definition of traffic
schedules, and rules for aircraft conflict resolution. TAAM
simulates the actual movement of all aircraft through the air-
space, collecting data on sequencing, delays, and potential
airborne conflicts. A simplistic ES is used to resolve potential
conflict between the simulated aircraft (24).
ATCoach, developed by UFA, Inc., and TRACON Pro, de-
veloped by Wesson International, are two examples of an off-
the-shelf simulation system designed to provide a comprehen-
sive training for air traffic controllers. Both systems include
elaborate simulation with user-friendly interface, including
speech capability and realistic graphics. The knowledge base
of the ATC operation is used to support the training compo-
nent. In ATCoach, the focus of ES is on monitoring the train-
ing session and providing domain guidance to the student
(25). In TRACON Pro, the AI techniques support planning of
airspace events using procedural techniques as researched by
Wesson (26).
EXPERT SYSTEMS IN AVIATION AND AEROSPACE SYSTEMS
Model-based reasoning is another AI approach, which bases
system behavior on the behavior of the system subcompo-

nents as represented by the frame-based model. Knowledge-
based autonomous test engineer (KATE), developed for the
National Aeronautics and Space Administration (NASA) by
Boeing Space Operations (27), is a generic software shell for
performing model-based monitoring, fault detection, diagno-
sis, and control. The four subsystems are (1) simulation, (2)
monitoring, (3) diagnosis, and (4) control. The system origi-
nated in the mid 1980s as a tool to support the operation of
the launch processing system. KATE was particularly de-
signed to check sensor operation for the Space Shuttle liquid-
oxygen loading system. The system is based on a model of the
sensor structure and diagnoses sensor failures. By separation
of the system structure from the component functions, a more
generic tool was designed. During the early 1990s, the system
started its operational application monitoring the tanking
data. The system was redesigned and implemented in C⫹⫹
programming language using popular Motif windowing envi-
ronment on a UNIX workstation to serve as a part of the vehi-
cle health management system.
Yet another facet of ES application is in the area of plan-
ning and scheduling. One example of such application is the
automatic cockpit crew scheduling developed by Japan Air-
lines and NEC (28) The system is designed to prepare
monthly schedules for flight crews. The system knowledge is
represented in frames and rules. The system’s distributed ar-
chitecture allows it to run inferencing on slave computers,
with the master computer serving as a cooperative inference
area and the monitor of data integrity. The backtracking tech-
nique is used to break a deadlock when the crew assignment
can not be found. Another example is an ES tool to support
shift duty assignments for an airport staff (29). The rules-
based system combines forward-chaining inference and con-
straints-relaxation techniques. It produces a timetable start-
ing with the initial assignment and continuing through the
iterative improvement process. The prototype has been tested
in airport operations.
FUTURE TRENDS
ES already plays a vital role in the safety and effectiveness
of complex systems. Their future in aerospace includes auton-
omous vehicles in both military and passenger aircraft; coop-
erating ES, such as those that would provide separation of
aircraft in flight; ATC systems that improve the safety and
efficiency of airspace use and of airports; and, to some extent,
training systems that deliver individualized lessons to stu-
dents.
The need to capture knowledge regarding the human oper-
ator in the aerospace system is clear; however, the ability to
accurately and effectively describe that knowledge in today’s
complex systems is becoming less practical using old tech-
niques. The future of ES design will focus on practical knowl-
edge engineering techniques that use the target system as a
means of collecting information and creating knowledge about
the users. In such a systemic approach, knowledge engi-
neering will evolve to include knowledge about the human,
human systems interfaces, and the systemic effects on human
operators interpretation of the system feedback. The use of
such developing technologies as neural networks and ES that
adapt will be more prominent than in systems in use today.
The adaptive system is capable of both induction and adapta-
AEROSPACE EXPERT SYSTEMS 311
tion—principles of elective automation that expose the need
for the designer to consider the individual variance of each
potential system operator—thereby providing a system that
can have various (infinite) combinations of automated
support.
As the ES designer elicits knowledge to support the design
process, the use of new allocation techniques will arise. This
new ES design approach is driven primarily by the need to
resolve a design conflict involving the expertise levels of the
user. Experts exhibit more bias than novices, declarative
knowledge forms a better foundation for self-critique, and ex-
perts adopt practices in lieu of formal procedures. This shift
in the user’s mental attitude creates a need to build an ES
that will allow the user to transition from novice to expert
over time, while recognizing the shift in bias based on decay
of formal knowledge (declarative and procedural) in favor of
experience and practices. Failure to allocate functions cor-
rectly in this transition-capable system is a breech of the de-
sign principle of flexibility in elective automation and will re-
sult in substandard human–system performance as both
levels.
The need to build transitory systems stems from the com-
puter community itself similar to adaptive design principles.
The desire to build small-scale systems has existed for de-
cades; however, practical aspects of building a microsystem,
such as complexity of design, cost, and production engineering
have been seen as limitations. The microprocessor-facilitated
size reduction of computing equipment to a level that made
applications practical in typical human-system interfaces.
This revolution in size has allowed rapid (and sometimes
careless) implementation of technology into human–system
interfaces. These implementations have changed the way hu-
mans understand and interact with systems and, as the gen-
eration of users that has grown up without such aids leaves
the design workforce, system boundaries adjust to an artifi-
cial, and possibly arbitrary new set of limitations that con-
strain the next-generation designer. This is the foundation of
the need for sound knowledge engineering and functional al-
location.
New systems are capable of more and the assumption that
humans are therefore not capable or encouraged to do as
much in the system as in the past will gradually become a
pseudo-standard for designers. This will lead to the extinction
of human capabilities unless the system design is tempered
with an understanding that humans are capable of certain
tasks, desire to perform certain tasks, and that allocation of
these tasks to ES when humans are capable of them is chang-
ing the system dynamics. The correct allocation depends on a
number of factors including reliability, cost, and efficacy of
the allocation, preference, and systemic risk mitigation.
ES that are currently in use, such as those described in
this article, will tend to proliferate in the aerospace commu-
nity. Synergy among these ES is likely to lead to cooperating
ES that seek out and elicit knowledge from each other. This
introduces another growing area where the future holds a
paradox for designers and implementers. The need to assess
the efficacy of information are paramount to human decision
making. Primary factors affecting the efficacy of the informa-
tion is the medium, the source, and the timeliness of that
information. Relevance of the information to the problem at
hand enters as a secondary criterion when there is ample in-
formation pertaining to the problem. The abundance of infor-
312 AEROSPACE EXPERT SYSTEMS
mation (contrasted with the dearth of information), defines
the ends of the spectrum in the paradox. The designers job is
to foresee the sources, define the timeliness attributes, and
deliver the information to the system inference engine so that
the process will continue with a high level of confidence and,
yet within a reasonable period of time as defined by the
knowledge engineers and the users.
At this point it becomes obvious that the designer has set
about controlling the ES process and, thus, has limited the
system to the boundaries of efficacy that the designer(s) fore-
see as reasonable. To overcome this contradiction, the ES of
the future will use a combination of Bayesian and induction
processes to match heuristic models and specific knowledge
seeking to reach theoretical absolute certainty, while settling
for something less to ensure that the process continues. Lim-
its of time will be systemically controlled using system-de-
rived rules. For example, an ES can learn to tell time due to
the ratio quality of the data when these data are input over
a considerable amount of time. ES have high confidence in
the longitudinal approach to using ratio data in developing
knowledge. ES are considered to be reasoning when either the
data type or the duration of the data input is less than contin-
uous and extended. The use of quantitative data can improve
the use of short-duration inputs, but the less preferred data
types (nominal, ordinal, and to some extent interval) create
certainty problems for today’s ES. This is likely to be the next
major improvement in expert systems in aerospace applica-
tions—the ability for the system to reason using data that
are meaningful to human processing, but not yet reliable for
computing.
BIBLIOGRAPHY
1. M. S. Nolan, Fundamentals of Air Traffic Control, Belmont, CA:
Wadsworth 1993.
2. A. L. Elias and J. D. Pararas, Potential use of artificial intelli-
gence techniques in air traffic control, Transportation Research
Circular, TRB, National Research Council, Washington, DC, AI
Workshop Report, 1985, pp. 17–31.
3. Webster’s New World Collegiate Dictionary, 3rd ed. New York:
Simon & Schuster, 1996.
3a. S. Haykin, Neural Networks: A Comprehensive Foundation, New
York: Macmillan, 1994; A. P. Sage (ed.), Concise Encyclopedia of
Information Processing in Systems and Organizations, New York:
Pergamon, 1990.
4. L. L. Smith, The distributed intelligence system and aircraft pi-
lotage, in AI and Simulation, San Diego: Simulation Councils
Inc., SCS, 1985, pp. 26–28.
5. A. Kornecki, Simulation and AI in aviation training, in G. W.
Zobrist and J. V. Leonard (eds.), Progress in Simulation, vol 2,
New York: Ablex Publishing, 1994, pp. 91–122.
6. J. G. Moser, Integration of AI and simulation in a comprehensive
decision-support system, Simulation, 47 (6): 223–229.
7. Y. Nakatani, M. Tsukiyama, and T. Fukuda, Case-based reason-
ing and decision aid for engineering design, in Proceedings of the
World Congress on Expert Systems, New York: Pergamon Press,
1991, pp. 369–376.
8. R. O. Duda, P. E. Hart, and N. Nilsson, Subjective Bayesian
methods for rule-based inference systems, in Proceedings of Na-
tional Computer Conference, AFIPS, 1976, pp. 1075–1082.
9. L. Zadeh, The role of fuzzy logic in the management of uncer-
tainty in expert systems, Fuzzy Sets Syst. (11), 199–227, 1983.
10. A. Kornecki, Artificial intelligence for air traffic, IEEE Potentials,
13 (3), 11–14, 1994.
11. A. Kornecki, SMARTFLO—Knowledge based system to support
central flow operations, in Proceedings of the 37th Annual Air
Traffic Control Association Conference, 1992, pp. 862–869.
12. T. Diefenbach, D. Carl, and M. Towhidnajad, Intelligent tutoring
and air traffic control training, in Proceedings of the 37th Air Traf-
fic Control Association Conference, 1992, pp. 489–494.
13. A. Kornecki et al., Intelligent tutoring issues for ATC training
system, IEEE Trans. Contr. Syst. Technol., 1 (3): 204–211, 1993.
14. V. Galotti and A. Kornecki, Knowledge engineering for an air
traffic expert system, in Proceedings of the 36th Annual Air Traffic
Control Association Conference, 1991, pp. 207–211.
15. J. Giarratano and G. Riley, Expert Systems, Principles and Pro-
gramming, Boston: PWS-Kent Publishing, 1989.
16. A. Kornecki, Building of an air traffic control expert system, in
M. Hamza (ed.),Proceedings of the IASTED International Sympo-
sium on ES, Anaheim, CA: Acta Press, 1989, pp. 217–219.
17. J. M. Hammer and R. Small, Intelligent interface in an associate
system, in W. B. Rouse (ed.), Human Technology Interaction in
Complex Systems, Greenwich, CT: JAI Press, 1995.
18. E. J. Bass, S. Ernst-Fortin, and R. Small, Knowledge base devel-
opment tool requirements for an intelligent monitoring aid, in
Proceedings of the 10th FLAIRS, 1997, pp. 412–416.
19. D. Snyder et al., Knowledge acquisition of tactical air-to-ground
mission information using system mapping, in Proceedings of Na-
tional Aerospace and Electronics Conference, New York, IEEE,
1992, pp. 668–674.
20. M. R. Anderson, C. Clark, and G. Dungan, Flight test maneuver
design using a skill- and rule-based pilot model, in Proceedings
of International Conference on Systems, Man and Cybernetics, New
York: IEEE, 1995, pp. 2682–2687.
21. P. Fishwick, Simulation Model Design and Execution—Building
Digital Worlds, Englewood Cliffs, NJ: Prentice Hall, 1995.
22. R. O’Keefe, Simulation and expert systems—a taxonomy and
some examples, Simulation, 46 (1): 10–16, 1986.
23. A. de Albornoz and F. E. Cellier, Building intelligence into an
autopilot—using qualitative simulation to support global decision
making, Simulation, 62: 354–363, 1994.
24. Anonymous, TAAM—Total Airport Airspace Modeller Product
Profile, Richmond, Victoria, Australia: The Preston Group, 1991.
25. J. A. Scardina, P. Y. Ryberg, and A. Gerstenfeld, Future ATC
automation aids based upon AI technology, Proc. IEEE, 77: 1625–
1633, 1989.
26. R. B. Wesson, Problem solving in the world of an air traffic con-
troller, doctoral dissertation, UT&A, University Microfilms Inter-
national, Ann Arbor, MI, December 1977.
27. C. O. Pepe et al., KATE—a project overview and software de-
scription, report, Boeing Aerospace Operations, Kennedy Space
Center, FL, March 1992.
28. K. Onodera and A. Mori, Cockpit crew scheduling and supporting
system, in Proceedings of the World Congress on Expert Systems,
New York: Pergamon Press, 1991, pp. 1–10.
29. K. P. Chow and C. K. Hui, Knowledge based approach to airport
staff rostering: a case study, in Proceedings of the World Congress
on Expert Systems, New York: Pergamon Press, 1991, pp. 46–53.
Reading List
J. Liebowitz, Introduction to Expert Systems, Santa Cruz, CA: Mitch-
ell, 1988.
E. Wagner, Artificial Intelligence and Tutoring Systems, Los Altos,
CA: Morgan Kaufman, 1987.
 AEROSPACE SIMULATION 313

T. I. Oren, Artificial intelligence and simulation, AI Applied to Simu-

lation, 18 (1), 3–8, 1986.
A. M. Wildberger, Integrating an expert system component into a
simulation, AI Papers, 20 (1), 132–135, 1988.
R. H. Michaelsen, D. Michie, and A. Boulanger, The technology of
expert systems, BYTE, 10: 303–312, April 1985.
F. Hayes-Roth, P. Klahr, and D. J. Mostow, Knowledge acquisition,
knowledge programming, and knowledge refinement, in P.Klahr
(ed.), The Rand Corporation, R-2540-NSF, 1980, Reading, MA: Ad-
dison-Wesley, 1986, pp. 310–349.
A. Gerstenfeld, Simulation combined with cooperating expert sys-
tems: an aid for training, screening, plans and procedures, J. ATC,
30: 33–35, 1988.
D. Spencer, Development environment for an ATC expert system, in
Transportation Research Circular, TRB, National Research Coun-
cil, Washington, DC, AI Workshop Report, 1985, pp. 32–37.
C. A. Shively, AIRPACK: Advisor for intelligent resolution of pre-
dicted aircraft conflicts, Transportation Research Circular, TRB,
National Research Council, Washington, DC, AI Workshop Re-
port, 1985, pp. 58–64.
A. Gonzalez et al., Simulation based expert system for training air
traffic controllers, in M. B. Fishman (ed.), Advances in Artificial
Intelligence Research, Greenwich, CT: JAI Press, 1989, pp.
295–308.
R. Steeb et al., Distributed problem solving for air fleet control:
framework and implementation, in P. Klahr (ed.), The Rand Cor-
poration, N-2139-ARPA, 1984, Reading, MA: Addison-Wesley,
1986, pp. 391–432.
P. McKinnon, Living with artificial intelligence, J. ATC, 29: 23–25,
1987.

Web Sites
Federal Aviation Administration http://www.faa.gov
MITRE Center for Advanced Aviation System Development http://
www.caasd.org
National Aeronautics and Space Agency Ames Research Center—
Advanced Air Transportation Technologies http://aatt.arc.nasa.gov
Massachussets Institute of Technology—Lincoln Laboratory http://
www.ll.mit.edu
AI resources http://www.cs.reading.ac.uk/people/dwc/ai.html

ANDREW J. KORNECKI
JAMES W. BLANCHARD
Embry-Riddle Aeronautical
University

AEROSPACE INDUSTRY. See AIR TRAFFIC.